Time domain classification and quantification of seismic noise in an urban environment

P>Broad-band urban seismic noise (USN) must be considered as a temporally and spatially non-stationary random process. Due to the high variability of USN a single measure like the standard deviation of a seismic noise time-series or the power spectral density at a given frequency is not enough to characterize a sample (time-series) of USN comprehensively. Therefore, we use long-term spectrograms and propose an automated statistical classification in the time domain to quantify and characterize USN. Long-term spectrograms of up to 28 d duration are calculated from a broad-band seismic data set recorded in the metropolitan area of Bucharest, Romania, to identify the frequency-dependent behaviour of the time-variable processes contributing to USN. Based on the spectral analysis eight frequency ranges between 8 mHz and 45 Hz are selected for our proposed time domain classification. The classification scheme identifies deviations from the Gaussian distribution of 4-hr-long time-series of USN. Our classification is capable to identify Gaussian distributed seismic noise time-series as well as time-series dominated by transient or periodic signals using six noise classes. Four additional noise classes are introduced to identify corrupt time-series. The performance of the method is tested with a synthetic data set. We also apply the statistical classification for the data set from Bucharest in three time windows (0-4, 8-12 and 13-17 EET) at 11 d in the eight frequency ranges. Only 40 per cent of the analysed time-series are observed to be Gaussian distributed. Most common deviations from the Gaussian distribution (similar to 47 per cent) are due to the influence of large-amplitude transient signals. In all frequency ranges between 0.04 and 45 Hz significant variations of the statistical properties of USN are observed with daytime, indicating the broad-band human influence on USN. We observe the human activity as a dominant influence on the USN above and below the frequency band of ocean-generated microseism between 0.04 and 0.6 Hz. Our time domain classification and quantification is furthermore capable to resolve the influence of wind on seismic noise and a known site effect variation in the metropolitan area of Bucharest. The information about noise amplitudes and statistical properties derived automatically from broad-band seismic data can be used to select time windows containing adequate data for seismic noise utilization like H/V-studies or ambient noise tomography.